Metric Calibration Dashboard

Enable interactive configuration and validation of risk thresholds and fairness metrics for organizational context

42

Active Metrics

38

Calibrated

4

Pending Calibration

2 hours ago

Last Calibration

Metric Threshold Calibration

Demographic Parity Difference

Measures the difference in positive outcome rates between demographic groups

Threshold Configuration

Green Threshold: 95%

Amber Threshold: 85%

Current Performance

Models in Green	142
Models in Amber	87
Models in Red	18

Formula: |P(Ŷ=1|D=privileged) - P(Ŷ=1|D=unprivileged)| < threshold

Calibration History & Trends

Threshold Evolution

Green Threshold

Amber Threshold

Model Distribution Changes

Weighted Risk Score Configuration

Configure the weights for each risk dimension in the unified AI Risk Score calculation

AI Risk Score (ARS) = Σ(w_i × s_i) where w_i is the weight and s_i is the normalized score

Fairness Bias:

15%

Security:

12%

Drift:

10%

Opaqueness:

Robustness:

10%

Privacy:

12%

Performance:

Compliance:

15%

Operational:

Value:

Total Weight: 100%

Valid Configuration

Recent Calibration Actions

Timestamp	Metric	Action	Previous Value	New Value	User	Impact
2024-01-15 14:30	Demographic Parity	Threshold Update	0.08	0.05	Sarah Chen	23 models affected
2024-01-15 11:15	PSI	Weight Adjustment	12%	10%	Mike Johnson	Portfolio-wide
2024-01-14 16:45	Adversarial Score	Formula Change	v1.2	v1.3	AI Gov Team	178 models affected
2024-01-14 09:30	SHAP Fidelity	Threshold Update	80%	85%	Lisa Wang	12 models affected

AI-Powered Calibration Recommendations

Demographic Parity

High Priority

Tighten threshold from 0.05 to 0.03

Based on: 95% of models already meet stricter threshold, regulatory trend toward tighter fairness requirements

Data Drift PSI

Medium Priority

Increase monitoring frequency to hourly

Based on: 3 critical models showed rapid drift in past week, early detection would prevent issues

Security Score Weight

High Priority

Increase weight from 12% to 18%

Based on: Recent industry attacks highlight security importance, peer benchmarks show 15-20% typical

Explainability Threshold

Low Priority

Consider model-specific thresholds

Based on: High-risk models need 90%+ fidelity, low-risk could accept 70%